The paper introduces ImpactBench, an open benchmark created to measure how generative AI affects the people who use it. Rather than focusing on a single metric, the platform is designed to capture the multi-dimensional impact of these models on individuals, from practical outcomes to broader personal effects.
ImpactBench is described as expert-guided, meaning its evaluation framework is shaped by expert input to ensure relevance and depth. The authors position it as a holistic tool, one that can be applied at scale to assess AI's influence across different contexts and user groups.
The work addresses a growing need for standardized, transparent ways to evaluate AI beyond technical performance. By making the benchmark open, the researchers aim to support broader adoption and continued refinement of impact measurement in the field.